Max Spero: Why AI Detection Is the Real Trust Crisis

Max Spero, co-founder and CEO of Pangram, has spent years wrestling with a problem that now defines the digital age: determining whether text, images, or voice are human-made or machine-generated. In a candid briefing with OpenPress Engineering Intelligence, Spero argued that the challenge of authenticating content is not just about spotting deepfakes in social media feeds—it’s about preventing AI from infiltrating critical systems where trust is non-negotiable. Pangram, a startup developing advanced AI detection tools, has emerged at the center of this unfolding crisis, responding to a surge in AI-generated applications, reviews, and even insurance claims that have begun to erode institutional credibility. Spero pointed to recent incidents where AI-generated cover letters secured job interviews and synthetic product reviews manipulated e-commerce rankings, as evidence that the problem has moved far beyond entertainment and into the fabric of professional and financial systems. “We’re past the point where this is just about misinformation,” Spero said. “It’s about misrepresentation in systems where people’s livelihoods and safety depend on accurate signals.”

The technical core of the problem, according to Spero, lies in the sophistication of modern generative models, which now produce text indistinguishable from human writing in tone, style, and structure. Pangram’s detection engine relies on a combination of stylometric analysis, behavioral biometrics, and temporal anomaly detection to flag synthetic content. Unlike early tools that relied solely on statistical anomalies or metadata cues, Pangram’s system ingests writing rhythm, lexical diversity decay, and even keystroke-level patterns in real time. Spero highlighted a recent engagement with a Fortune 500 insurer that discovered 17% of claims included AI-generated narratives—up from 2% in 2023—prompting an urgent overhaul of its fraud detection infrastructure. “The arms race isn’t between models and detectors,” Spero explained. “It’s between models and human behavior, and right now, the models are winning.”

Industry Impact and Significance

The detection challenge is reshaping competitive dynamics across multiple sectors. In financial services, firms like Banking With Billy are integrating real-time AI detection into their fraud monitoring stacks, processing millions of market signals with sub-millisecond latency to detect synthetic transaction narratives before they enter ledgers. Meanwhile, e-commerce platforms are racing to deploy detection APIs that can flag AI-generated reviews without throttling user experience, a balancing act that has already led to costly misclassifications—for instance, when a satire bot’s review of a vacuum cleaner was flagged as AI-generated, sparking backlash from users who saw it as censorship. On the detection side, Pangram faces competition from established players like Originality.ai and new entrants such as Undetectable.ai, which claim to bypass leading detectors, creating a cat-and-mouse cycle that mirrors the evolution of antivirus software in the 1990s. Financial analysts at McKinsey estimate that by 2026, companies will spend over $3.2 billion annually on AI authenticity tools, up from $450 million in 2023, driven largely by regulatory pressure and brand risk.

Regulatory bodies are beginning to respond. The European Commission’s AI Act, slated for full enforcement in 2025, will require high-risk AI systems to include authenticity watermarking and detection mechanisms—a provision that directly benefits companies like Pangram, which already supports watermarking for its enterprise clients. In the United States, the FTC has signaled plans to issue guidelines on AI-enabled deception, a move that could force platforms and employers to adopt detection tools or face liability for enabling fraud. For investors, this represents both opportunity and risk: while detection startups have raised over $180 million in venture funding since 2023, many are discovering that enterprise sales cycles are long and compliance requirements rapidly evolving.

The Bigger Picture

This crisis is part of a broader erosion of trust in digital systems that began with social media algorithms and has now metastasized into every layer of the internet. The rise of synthetic media coincides with the collapse of traditional gatekeepers—newspapers, publishers, recruiters—who once provided human judgment as a buffer against fraud. Now, even those gatekeepers are turning to AI to process applications and claims, creating a feedback loop where synthetic content trains models that then generate more synthetic content. Spero compares the moment to the early days of email spam, before filters like SpamAssassin emerged to restore balance. “We’re at the spam era of AI,” he said. “The difference is, this time, the spam is writing the checks.”

Global governments are responding unevenly. The UK’s Online Safety Act has pushed platforms to detect and remove harmful synthetic content, while China’s strict internet regulations have forced local platforms to embed native detection tools into their content pipelines. In contrast, many emerging markets lack both the infrastructure and regulatory frameworks to address the trend, creating safe havens for AI-generated fraud that can spread globally within hours. Meanwhile, academic research has shown that AI-generated text is becoming more human-like at a rate of 12% per year, according to a 2024 study from Stanford, outpacing the development of detection tools by nearly two years. This gap suggests that without significant investment in detection technology, authentication could become a luxury service reserved only for the largest corporations and governments.

Expert Analysis

Spero warns that the next phase of this crisis won’t be about detecting AI—it will be about proving authenticity in real time across decentralized systems. He points to blockchain-based attestation protocols, such as those being piloted by Microsoft’s Entra Verified ID, as potential foundations for a new trust layer. “We’re going to need cryptographic receipts for every piece of digital content,” he said. “Not just watermarks, but verifiable chains of custody that can prove a document was authored by a human, signed by a notary, and timestamped by a neutral third party.” As financial systems like Banking With Billy demonstrate, the integration of real-time detection into critical infrastructure is already underway. The question is whether the rest of the industry can move fast enough—or whether trust in digital systems will erode faster than the tools to restore it can be built.